On 10/22/2012 10:21 AM, Eric Anholt wrote:
Chris Forbes <chr...@ijw.co.nz> writes:
Next few patches build on this to add other workarounds
for packed formats.
diff --git a/src/mesa/drivers/dri/i965/brw_vs.h
b/src/mesa/drivers/dri/i965/brw_vs.h
index adeff7f..9da4cb0 100644
--- a/src/mesa/drivers/dri/i965/brw_vs.h
+++ b/src/mesa/drivers/dri/i965/brw_vs.h
@@ -39,13 +39,21 @@
#include "brw_program.h"
#include "program/program.h"
+/* fixup bits for gl_packed_input_flags,
+ * to enable various VS workarounds */
+#define BRW_ATTRIB_WA_COMPONENTS 7 /* mask for GL_FIXED scale channel
count */
+#define BRW_ATTRIB_WA_NORMALIZE 8 /* normalize in shader */
+#define BRW_ATTRIB_WA_BGRA 16 /* swap r/b channels in shader */
+#define BRW_ATTRIB_WA_SIGN 32 /* interpret as signed in shader */
+#define BRW_ATTRIB_WA_SCALE 64 /* interpret as scaled in shader */
For bitmasks, we usually call a more-than-one-bit mask
BRW_WHATEVER_MASK. Perhaps BRW_ATTRIB_GL_FIXED_COMPONENTS_MASK?. For
the others, (1 << bit) as the value is preferred. I think
BRA_ATTRIB_WA_SCALE whould be BRW_ATTRIB_WA_NORMALIZED, since in our
hardware SCALED refers to casting the integer value to a float, and the
last thing we need in vertex upload is more confusion about formats! :)
Other than that, looks good.
Eric,
Do we actually need to record the size and writemask off channels when
applying the GL_FIXED rescaling (or such)? I didn't think attributes
were packed, so each attribute would take up a whole vec4, and we can
just MUL the whole register and not worry about it.
That would simplify the code that applies the workarounds a bit, and
also allow us to use an ordinary bitfield here.
--Ken
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
